memory-management（内存管理）

[IMPORTANT] Use TaskCreate to break ALL work into small tasks BEFORE starting — including tasks for each file read. This prevents context loss from long files. For simple tasks, AI MUST ATTENTION ask user whether to skip.

Critical Thinking Mindset — Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence >80% to act. Anti-hallucination: Never present guess as fact — cite sources for every claim, admit uncertainty freely, self-check output for errors, cross-reference independently, stay skeptical of own confidence — certainty without evidence root of all hallucination.

AI Mistake Prevention — Failure modes to avoid on every task: - Check downstream references before deleting. Deleting components causes documentation and code staleness cascades. Map all referencing files before removal. - Verify AI-generated content against actual code. AI hallucinates APIs, class names, and method signatures. Always grep to confirm existence before documenting or referencing. - Trace full dependency chain after edits. Changing a definition misses downstream variables and consumers derived from it. Always trace the full chain. - Trace ALL code paths when verifying correctness. Confirming code exists is not confirming it executes. Always trace early exits, error branches, and conditional skips — not just happy path. - When debugging, ask "whose responsibility?" before fixing. Trace whether bug is in caller (wrong data) or callee (wrong handling). Fix at responsible layer — never patch symptom site. - Assume existing values are intentional — ask WHY before changing. Before changing any constant, limit, flag, or pattern: read comments, check git blame, examine surrounding code. - Verify ALL affected outputs, not just the first. Changes touching multiple stacks require verifying EVERY output. One green check is not all green checks. - Holistic-first debugging — resist nearest-attention trap. When investigating any failure, list EVERY precondition first (config, env vars, DB names, endpoints, DI registrations, data preconditions), then verify each against evidence before forming any code-layer hypothesis. - Surgical changes — apply the diff test. Bug fix: every changed line must trace directly to the bug. Don't restyle or improve adjacent code. Enhancement task: implement improvements AND announce them explicitly. - Surface ambiguity before coding — don't pick silently. If request has multiple interpretations, present each with effort estimate and ask. Never assume all-records, file-based, or more complex path.

Quick Summary

Goal: Persist patterns, decisions, and task progress across sessions using two complementary memory systems.

Workflow:

1. File Checkpoints — Save task-specific context to plans/reports/checkpoint-*.md every 30-60 min
2. MCP Memory Graph — Store reusable knowledge (patterns, decisions, bug fixes) as typed entities with relations
3. Recovery — On context loss, find latest checkpoint via Glob, read it, resume from documented next steps

Key Rules:

• Use file checkpoints for task-specific progress; MCP memory for cross-session knowledge
• Create checkpoints before expected context compaction and at key milestones
• Always include Recovery Instructions in checkpoint files

Be skeptical. Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence percentages (Idea should be more than 80%).

Memory Management & Knowledge Persistence

Build and maintain a knowledge graph of patterns, decisions, and learnings across sessions. Also provides external file-based checkpoints for long-running tasks.

Two Memory Systems

Use MCP Memory for reusable knowledge. Use File Checkpoints for task-specific context.

Part 1: File-Based External Memory (Checkpoints)

When to Create File Checkpoints

• Starting complex multi-step tasks (investigation, planning, implementation)
• Every 30-60 minutes during long tasks
• At key milestones
• Before expected context compaction
• After completing significant analysis phases

Checkpoint File Location

Files saved to: plans/reports/checkpoint-{timestamp}-{slug}.md

CHECKPOINT_CREATE Protocol

Create a checkpoint file with this structure:

# Memory Checkpoint: {Task Description}

> Created: {ISO timestamp}
> Task Type: {investigation|planning|bugfix|feature|docs}
> Phase: {current phase number/name}

## Task Context

{What you're working on and why}

## Key Findings

{Critical discoveries and insights - be specific with file paths and line numbers}

## Files Analyzed

| File              | Purpose     | Status   |
| ----------------- | ----------- | -------- |
| path/file.cs:line | description | ✅/🔄/⏳ |

## Progress

- [x] Completed items
- [ ] In-progress items
- [ ] Remaining items

## Important Context

{Information that must be preserved - decisions, assumptions, rationale}

## Next Steps

1. {Immediate next action}
2. {Following action}

## Recovery Instructions

{Exact steps to resume: which file to read, which line to continue from}

CHECKPOINT_RECOVER Protocol

When recovering from a checkpoint:

1. Search for latest checkpoint: Glob("plans/reports/checkpoint-*.md")
2. Read the checkpoint file
3. Load any referenced analysis files
4. Review Progress section
5. Continue from documented Next Steps
6. Create new checkpoint after resuming

Auto-Checkpoint (PreCompact Hook)

The system automatically creates checkpoints before context compaction. These auto-checkpoints are minimal - for better context preservation, create manual checkpoints using /checkpoint.

Part 2: MCP Memory Graph (Knowledge Persistence)

Memory Entity Types

Memory Operations

Create New Entity

mcp__memory__create_entities([
    {
        name: 'EmployeeValidationPattern',
        entityType: 'Pattern',
        observations: [
            'Use project validation fluent API (see docs/project-reference/backend-patterns-reference.md)',
            'Chain with .And() and .AndAsync()',
            "Return validation result, don't throw",
            'Location: {Service}.Application/UseCaseCommands/'
        ]
    }
]);

Create Relationships

mcp__memory__create_relations([
    {
        from: 'ServiceA',
        to: 'ServiceB',
        relationType: 'depends_on'
    },
    {
        from: 'EmployeeEntity',
        to: 'UserEntity',
        relationType: 'syncs_from'
    }
]);

Add Observations

mcp__memory__add_observations([
    {
        entityName: 'EmployeeValidationPattern',
        contents: [
            'Also supports .AndNot() for negative validation',
            'Use .Of<ICqrsRequest>() for type conversion (see docs/project-reference/backend-patterns-reference.md)'
        ]
    }
]);

Search Knowledge

// Search by query
mcp__memory__search_nodes({ query: 'validation pattern' });

// Open specific entities
mcp__memory__open_nodes({
    names: ['EmployeeValidationPattern', 'ServiceAModule']
});

// Read entire graph
mcp__memory__read_graph();

Delete Outdated Knowledge

// Delete entities
mcp__memory__delete_entities({ entityNames: ['OutdatedPattern'] });

// Delete specific observations
mcp__memory__delete_observations([
    {
        entityName: 'EmployeeValidationPattern',
        observations: ['Outdated observation text']
    }
]);

// Delete relations
mcp__memory__delete_relations([
    {
        from: 'OldService',
        to: 'NewService',
        relationType: 'depends_on'
    }
]);

When to Save to Memory

Always Save

1. Discovered Patterns: New code patterns not in documentation
2. Bug Solutions: Complex bugs with non-obvious solutions
3. Service Boundaries: Which service owns what
4. Architectural Decisions: Why a particular approach was chosen
5. Anti-Patterns: Mistakes to avoid

Save at Session End

// Session summary template
mcp__memory__create_entities([
    {
        name: `Session_${taskName}_${date}`,
        entityType: 'SessionSummary',
        observations: [
            `Task: ${taskDescription}`,
            `Completed: ${completedItems.join(', ')}`,
            `Remaining: ${remainingItems.join(', ')}`,
            `Key Files: ${keyFiles.join(', ')}`,
            `Discoveries: ${discoveries.join(', ')}`,
            `Next Steps: ${nextSteps.join(', ')}`
        ]
    }
]);

Memory Retrieval Patterns

Session Start Protocol

// 1. Search for related context
const results = mcp__memory__search_nodes({
    query: 'current feature or task keywords'
});

// 2. Load relevant entities
mcp__memory__open_nodes({
    names: results.entities.map(e => e.name)
});

// 3. Check for incomplete sessions
mcp__memory__search_nodes({ query: 'SessionSummary Remaining' });

Before Implementation

// Check for existing patterns
mcp__memory__search_nodes({ query: 'CQRS command pattern' });

// Check for anti-patterns
mcp__memory__search_nodes({ query: 'AntiPattern command' });

// Check for related decisions
mcp__memory__search_nodes({ query: 'Decision validation' });

After Bug Fix

// Save the fix
mcp__memory__create_entities([
    {
        name: `BugFix_${bugName}`,
        entityType: 'BugFix',
        observations: [
            `Symptom: ${symptomDescription}`,
            `Root Cause: ${rootCause}`,
            `Solution: ${solution}`,
            `Files: ${affectedFiles.join(', ')}`,
            `Prevention: ${preventionTip}`
        ]
    }
]);

Knowledge Graph Structure

┌─────────────────────────────────────────────────────────────┐
│                     Project Knowledge                       │
├─────────────────────────────────────────────────────────────┤
│  Services                                                   │
│  ├── ServiceA ──depends_on──> AccountsService               │
│  ├── ServiceB ──depends_on──> AccountsService               │
│  └── ServiceC ──depends_on──> AccountsService               │
│                                                             │
│  Patterns                                                   │
│  ├── CQRSCommandPattern                                     │
│  ├── CQRSQueryPattern                                       │
│  ├── EntityEventPattern                                     │
│  └── ValidationPattern                                      │
│                                                             │
│  Entities                                                   │
│  ├── Employee ──syncs_from──> User                          │
│  ├── Company ──syncs_from──> Organization                   │
│  └── LeaveRequest ──owned_by──> ServiceA                     │
│                                                             │
│  Sessions                                                   │
│  ├── Session_LeaveRequest_2025-01-15                        │
│  └── Session_EmployeeImport_2025-01-14                      │
└─────────────────────────────────────────────────────────────┘

Importance Scoring

When saving observations, prioritize:

Memory Maintenance

Weekly Cleanup

// Find old session summaries (> 30 days)
mcp__memory__search_nodes({ query: 'SessionSummary' });

// Delete outdated sessions
mcp__memory__delete_entities({
    entityNames: ['Session_OldTask_2024-12-01']
});

Consolidation

When multiple observations cover same topic:

// 1. Read existing entity
mcp__memory__open_nodes({ names: ['PatternName'] });

// 2. Delete fragmented observations
mcp__memory__delete_observations([
    {
        entityName: 'PatternName',
        observations: ['Fragment 1', 'Fragment 2']
    }
]);

// 3. Add consolidated observation
mcp__memory__add_observations([
    {
        entityName: 'PatternName',
        contents: ['Consolidated comprehensive observation']
    }
]);

Quick Reference

Create: mcp__memory__create_entities / mcp__memory__create_relations Read: mcp__memory__read_graph / mcp__memory__open_nodes / mcp__memory__search_nodes Update: mcp__memory__add_observations Delete: mcp__memory__delete_entities / mcp__memory__delete_observations / mcp__memory__delete_relations

Part 3: Integration with Workflows

Long-Running Task Memory Pattern

All long-running workflows should follow this pattern:

┌─────────────────────────────────────────────────────────┐
│ TASK START                                               │
│   └── Create initial checkpoint with task context        │
│   └── Initialize todo list                               │
│                                                          │
│ EVERY 20-30 OPERATIONS                                   │
│   └── Update checkpoint with progress                    │
│   └── Update todo list status                            │
│                                                          │
│ MILESTONE REACHED                                         │
│   └── Create detailed checkpoint                         │
│   └── Save key findings to MCP memory (if reusable)      │
│                                                          │
│ BEFORE COMPACTION (auto via PreCompact hook)             │
│   └── Auto-checkpoint created by system                  │
│                                                          │
│ AFTER COMPACTION / SESSION RESUME                        │
│   └── Read latest checkpoint                             │
│   └── Search MCP memory for relevant context             │
│   └── Continue from documented Next Steps                │
│                                                          │
│ TASK COMPLETE                                             │
│   └── Final checkpoint with summary                      │
│   └── Save reusable patterns to MCP memory               │
│   └── Clean up temporary checkpoints                     │
└─────────────────────────────────────────────────────────┘

Checkpoint Naming Convention

Related Commands & Skills

Memory Decision Matrix

Related

• learn
• context-optimization

Closing Reminders

• MANDATORY IMPORTANT MUST ATTENTION break work into small todo tasks using TaskCreate BEFORE starting
• MANDATORY IMPORTANT MUST ATTENTION search codebase for 3+ similar patterns before creating new code
• MANDATORY IMPORTANT MUST ATTENTION cite file:line evidence for every claim (confidence >80% to act)
• MANDATORY IMPORTANT MUST ATTENTION add a final review todo task to verify work quality
• MUST ATTENTION apply critical thinking — every claim needs traced proof, confidence >80% to act. Anti-hallucination: never present guess as fact.
• MUST ATTENTION apply AI mistake prevention — holistic-first debugging, fix at responsible layer, surface ambiguity before coding, re-read files after compaction.

[TASK-PLANNING] Before acting, analyze task scope and systematically break it into small todo tasks and sub-tasks using TaskCreate.